CEPH extrem usage of RAM

[informatique-itm]

Hi everybody,

because of an unexpected crash of the proxmox backup VM, i checked the related HP server host
and noticed that SWAP space was used to 100%.

I ran swapoff -a and swapon -a, which freed the SWAP space.

Currently the 125GB RAM ares used up to 92%, but only 30% are used by running VM's (mainly the proxmox backup VM),
the rest are consumed by CEPH processes.
This consumption has been constantly growing over the last few months.

Is this 'normal' CEPH behaviour ?
Can i change that to avoid too much SWAP action ?

Regards,
Gaston

 

[datschlatscher]

As far as I am aware, processes do not have such fine control as to choose where their memory should be allocated. Swap usage is determined by something called swappiness, through which the Kernel decides how much of the SWAP it should be using. Sometimes though, the kernel might make misjudgements on how it uses the SWAP, leading it to use the SWAP for certain tasks even if that is unnecessary.
In such cases you can reduce the swappiness to a smaller number, which should reduce such unwanted behavior. You can also set it to 0, which tells the kernel to only swap when there is no RAM left in the system.

For more information for this setting please see this link: https://linuxhint.com/customize-linux-swappiness/

 

[jasonsansone]

Without knowing if this node is also a Ceph MDS, Manager, and how many OSD’s it has it is impossible to say how much memory Ceph should be consuming. However, Ceph like all SDS takes heavy advantage of memory caching.